Gemini 4 发布图表引发三家模型分工讨论
据称 Gemini 4 Argon 未发布先参测，去掉它后 Astra 在 19 项测试中赢 10 项，作者按场景给出选型建议。
BREAKING: Google's Gemini 4 launch chart just settled the ChatGPT vs Claude debate. Gemini 4 Argon isn't out yet. But Google tested it against GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1. Take Argon out and Astra wins 10 of the 19 tests, Opus 6 and Fable 2. Use GPT-6 Astra for: → long documents (71.8% vs 66.8% for Opus on 256K to 1M tokens) → charts and video (71.0% vs 66.3% on Chartography) → science and maths (85.4% vs 73.1% on LABBench 2) Use Claude Opus 5.5 for: → coding in the terminal (66.4% vs 58.2% on Terminal-bench 4.0) → general professional work (67.0% vs 65.8% for Fable on the Vals Index) → running business tasks end to end (42.5% vs 41.4% for Astra) Use Claude Fable 5.1 for: → financial research (58.9% vs 58.6% for Opus) → legal work, though none of them are good at it yet (6.7%, while Argon scores 19.6%) Bookmark it for the next time you're switching tabs.